Automatic detection of neovascularization on optic disk region with feature extraction and support vector machine

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Science

Background:

• Neovascularization (NV) is a critical indicator of Proliferative Diabetic Retinopathy (PDR), a condition that can cause sudden vision loss.
• The increasing prevalence of Diabetic Retinopathy (DR) outpaces the availability of ophthalmologists for timely screening.
• Automated detection of NV is crucial for efficient DR screening and preventing vision impairment.

Purpose of the Study:

• To develop and evaluate a computer-aided detection system for neovascularization in the optic disk region.
• To improve the efficiency and accessibility of screening for Proliferative Diabetic Retinopathy.

Main Methods:

• A novel image processing approach was employed, including vessel segmentation using multi-level Gabor filtering.
• Feature extraction involved analyzing vessel-related and texture features from retinal images.
• Machine learning algorithms were utilized for image classification of neovascularization.

Main Results:

• Twenty-one features were extracted from each image.
• The system was trained and tested on 66 retinal images (16 with NVD, 50 normal).
• Achieved a sensitivity of 15/16 and a specificity of 47/50 for detecting neovascularization.

Conclusions:

• The proposed computer-aided detection method shows significant potential for accurate identification of neovascularization in the optic disk.
• This technology can support automated Diabetic Retinopathy screening, addressing the shortage of specialists.
• Early and accurate detection of NV can help prevent vision loss in diabetic patients.